Abstract:
Vehicle detection and classification using deep learning methods has been found out in
this paper. In the area of highway management, vehicle detection and classification are
becoming more significant currently. Vehicle Detection and Classification based on
Multiple Deep Learning Methods has been found in this paper, multiple classes and
multiple methods have been used on this topic in very less research paper. In fact, there
are different types of vehicles, such as cars, microbuses, jeeps, pickups, buses, trucks,
taxis, vans, rickshaws, etc. Multiple vehicles have different shapes and sizes (bounding
boxes) so it is very difficult to detect this multiple class, in this paper multiple classes of
vehicle have been used. We have used three of the deep learning methods in this paper,
method performance, detection ability and object classification has been compared with
those methods. The three deep learning methods we have proposed are Mask R-CNN,
Faster R-CNN and Yolo V5 method. Here ResNet50 is used as backbone in Faster RCNN method and ResNet101 is used as backbone in Mask R-CNN method, where Mask
R-CNN and Faster R-CNN methods are included in CNN family ties. Though the Mask
R-CNN is the extension of Faster R-CNN. We evaluate our models' performance through
Confusion Matrix. The methods of F1 score, mean average recall and mean average
precision have been found out through the Confusion Matrix, the methods have been
compared with those values. From that value it is evident that Mask R-CNN gives better
performance than other methods. We see from the table (table: 6) that the following
values are obtained using Confusion Matrix from Mask R-CNN method F1 score - 87%,
mean average recall- 92% and mean average precision - 82%. So The Mask R-CNN's
detection score is higher than other models, so the Mask R-CNN's detection ability and
classification is better than other models.
There will be a lot of cooperation in vehicle detection and prediction for self-driving cars
or various robotic cars through this work.
